##// END OF EJS Templates
help: format revlog.txt more closely to result...
Martin von Zweigbergk -
r30827:e997e482 default
parent child Browse files
Show More
@@ -1,201 +1,210 b''
1 1 Revision logs - or *revlogs* - are an append only data structure for
2 2 storing discrete entries, or *revisions*. They are the primary storage
3 3 mechanism of repository data.
4 4
5 5 Revlogs effectively model a directed acyclic graph (DAG). Each node
6 6 has edges to 1 or 2 *parent* nodes. Each node contains metadata and
7 7 the raw value for that node.
8 8
9 9 Revlogs consist of entries which have metadata and revision data.
10 10 Metadata includes the hash of the revision's content, sizes, and
11 11 links to its *parent* entries. The collective metadata is referred
12 12 to as the *index* and the revision data is the *data*.
13 13
14 14 Revision data is stored as a series of compressed deltas against previous
15 15 revisions.
16 16
17 17 Revlogs are written in an append-only fashion. We never need to rewrite
18 18 a file to insert nor do we need to remove data. Rolling back in-progress
19 19 writes can be performed by truncating files. Read locks can be avoided
20 20 using simple techniques. This means that references to other data in
21 21 the same revlog *always* refer to a previous entry.
22 22
23 23 Revlogs can be modeled as 0-indexed arrays. The first revision is
24 24 revision #0 and the second is revision #1. The revision -1 is typically
25 25 used to mean *does not exist* or *not defined*.
26 26
27 27 File Format
28 28 ===========
29 29
30 30 A revlog begins with a 32-bit big endian integer holding version info
31 31 and feature flags. This integer is shared with the first revision
32 32 entry.
33 33
34 34 This integer is logically divided into 2 16-bit shorts. The least
35 35 significant half of the integer is the format/version short. The other
36 36 short holds feature flags that dictate behavior of the revlog.
37 37
38 38 Only 1 bit of the format/version short is currently used. Remaining
39 39 bits are reserved for future use.
40 40
41 41 The following values for the format/version short are defined:
42 42
43 43 0
44 44 The original revlog version.
45 45 1
46 46 RevlogNG (*next generation*). It replaced version 0 when it was
47 47 implemented in 2006.
48 48
49 49 The feature flags short consists of bit flags. Where 0 is the least
50 50 significant bit, the following bit offsets define flags:
51 51
52 52 0
53 53 Store revision data inline.
54 54 1
55 55 Generaldelta encoding.
56 56
57 57 2-15
58 58 Reserved for future use.
59 59
60 60 The following header values are common:
61 61
62 62 00 00 00 01
63 63 RevlogNG
64 64 00 01 00 01
65 65 RevlogNG + inline
66 66 00 02 00 01
67 67 RevlogNG + generaldelta
68 68 00 03 00 01
69 69 RevlogNG + inline + generaldelta
70 70
71 71 Following the 32-bit header is the remainder of the first index entry.
72 72 Following that are remaining *index* data. Inlined revision data is
73 73 possibly located between index entries. More on this layout is described
74 74 below.
75 75
76 76 RevlogNG Format
77 77 ===============
78 78
79 79 RevlogNG (version 1) begins with an index describing the revisions in
80 80 the revlog. If the ``inline`` flag is set, revision data is stored inline,
81 81 or between index entries (as opposed to in a separate container).
82 82
83 83 Each index entry is 64 bytes. The byte layout of each entry is as
84 84 follows, with byte 0 being the first byte (all data stored as big endian):
85 85
86 86 0-3 (4 bytes) (rev 0 only)
87 87 Revlog header
88
88 89 0-5 (6 bytes)
89 90 Absolute offset of revision data from beginning of revlog.
91
90 92 6-7 (2 bytes)
91 93 Bit flags impacting revision behavior. The following bit offsets define:
92 94 0: REVIDX_ISCENSORED revision has censor metadata, must be verified.
93 95 1: REVIDX_EXTSTORED revision data is stored externally.
96
94 97 8-11 (4 bytes)
95 98 Compressed length of revision data / chunk as stored in revlog.
99
96 100 12-15 (4 bytes)
97 101 Uncompressed length of revision data. This is the size of the full
98 102 revision data, not the size of the chunk post decompression.
103
99 104 16-19 (4 bytes)
100 105 Base or previous revision this revision's delta was produced against.
101 106 -1 means this revision holds full text (as opposed to a delta).
102 107 For generaldelta repos, this is the previous revision in the delta
103 108 chain. For non-generaldelta repos, this is the base or first
104 109 revision in the delta chain.
110
105 111 20-23 (4 bytes)
106 112 A revision this revision is *linked* to. This allows a revision in
107 113 one revlog to be forever associated with a revision in another
108 114 revlog. For example, a file's revlog may point to the changelog
109 115 revision that introduced it.
116
110 117 24-27 (4 bytes)
111 118 Revision of 1st parent. -1 indicates no parent.
119
112 120 28-31 (4 bytes)
113 121 Revision of 2nd parent. -1 indicates no 2nd parent.
122
114 123 32-63 (32 bytes)
115 124 Hash of revision's full text. Currently, SHA-1 is used and only
116 125 the first 20 bytes of this field are used. The rest of the bytes
117 126 are ignored and should be stored as \0.
118 127
119 128 If inline revision data is being stored, the compressed revision data
120 129 (of length from bytes offset 8-11 from the index entry) immediately
121 130 follows the index entry. There is no header on the revision data. There
122 131 is no padding between it and the index entries before and after.
123 132
124 133 If revision data is not inline, then raw revision data is stored in a
125 134 separate byte container. The offsets from bytes 0-5 and the compressed
126 135 length from bytes 8-11 define how to access this data.
127 136
128 137 The first 4 bytes of the revlog are shared between the revlog header
129 138 and the 6 byte absolute offset field from the first revlog entry.
130 139
131 140 Delta Chains
132 141 ============
133 142
134 143 Revision data is encoded as a chain of *chunks*. Each chain begins with
135 144 the compressed original full text for that revision. Each subsequent
136 145 *chunk* is a *delta* against the previous revision. We therefore call
137 146 these chains of chunks/deltas *delta chains*.
138 147
139 148 The full text for a revision is reconstructed by loading the original
140 149 full text for the base revision of a *delta chain* and then applying
141 150 *deltas* until the target revision is reconstructed.
142 151
143 152 *Delta chains* are limited in length so lookup time is bound. They are
144 153 limited to ~2x the length of the revision's data. The linear distance
145 154 between the base chunk and the final chunk is also limited so the
146 155 amount of read I/O to load all chunks in the delta chain is bound.
147 156
148 157 Deltas and delta chains are either computed against the previous
149 158 revision in the revlog or another revision (almost certainly one of
150 159 the parents of the revision). Historically, deltas were computed against
151 160 the previous revision. The *generaldelta* revlog feature flag (enabled
152 161 by default in Mercurial 3.7) activates the mode where deltas are
153 162 computed against an arbitrary revision (almost certainly a parent revision).
154 163
155 164 File Storage
156 165 ============
157 166
158 167 Revlogs logically consist of an index (metadata of entries) and
159 168 revision data. This data may be stored together in a single file or in
160 169 separate files. The mechanism used is indicated by the ``inline`` feature
161 170 flag on the revlog.
162 171
163 172 Mercurial's behavior is to use inline storage until a revlog reaches a
164 173 certain size, at which point it will be converted to non-inline. The
165 174 reason there is a size limit on inline storage is to establish an upper
166 175 bound on how much data must be read to load the index. It would be a waste
167 176 to read tens or hundreds of extra megabytes of data just to access the
168 177 index data.
169 178
170 179 The actual layout of revlog files on disk is governed by the repository's
171 180 *store format*. Typically, a ``.i`` file represents the index revlog
172 181 (possibly containing inline data) and a ``.d`` file holds the revision data.
173 182
174 183 Revision Entries
175 184 ================
176 185
177 186 Revision entries consist of an optional 1 byte header followed by an
178 187 encoding of the revision data. The headers are as follows:
179 188
180 189 \0 (0x00)
181 190 Revision data is the entirety of the entry, including this header.
182 191 u (0x75)
183 192 Raw revision data follows.
184 193 x (0x78)
185 194 zlib (RFC 1950) data.
186 195
187 196 The 0x78 value is actually the first byte of the zlib header (CMF byte).
188 197
189 198 Hash Computation
190 199 ================
191 200
192 201 The hash of the revision is stored in the index and is used both as a primary
193 202 key and for data integrity verification.
194 203
195 204 Currently, SHA-1 is the only supported hashing algorithm. To obtain the SHA-1
196 205 hash of a revision:
197 206
198 207 1. Hash the parent nodes
199 208 2. Hash the fulltext of the revision
200 209
201 210 The 20 byte node ids of the parents are fed into the hasher in ascending order.
General Comments 0
You need to be logged in to leave comments. Login now